Articles tagged with Artificial Neural Networks
Researchers at the University of Tokyo have developed a new technique to protect sensitive AI-based applications from attackers. By adding random noise to the inner layers of neural networks, they improved the resilience of these systems. This approach promotes greater adaptability and reduces susceptibility to simulated adversarial at...
Researchers at Osaka University develop a method to train AI models using simulated city images, reducing the need for real data and saving human effort. The approach generates realistic images with accurate ground truth labels, addressing instance segmentation challenges.
A new study by North Carolina State University found that artificial intelligence performs better when it chooses diversity over lack of diversity. The AI was able to increase its accuracy up to 10 times more than conventional AI in solving complicated problems.
Researchers from Okayama University used CNNs to estimate rice yield from pre-harvest photographs, achieving accuracy of 68-69%. The model highlighted the importance of panicles in yield estimation and showed promise for monitoring rice productivity at regional scales.
Researchers developed an AI tool called DECIMER that can translate chemical structural formulae into machine-readable codes. This allows for the automatic search and processing of scientific articles containing chemical information.
Researchers developed a neural network-based system, PAT, for snapshot compressive imaging. It achieves comparable image quality to CASSI and holds strong promise due to advancements in AI processing capabilities.
Researchers propose a hypothesis that astrocytes, non-neuronal cells in the brain, can perform core computation as transformers, providing insights into human brain function and machine learning success. This discovery could spark future neuroscience research and help explain transformer performance across complex tasks.
Researchers used a mathematical theory called the free energy principle to predict how real neural networks learn and organize themselves. The study successfully mimicked this process in rat embryo neurons grown in a culture dish, demonstrating the principle's guiding force behind biological neural network learning.
A new complex-domain neural network enhances large-scale coherent imaging by exploiting latent coupling information between amplitude and phase components. The technique reduces exposure time and data volume significantly while maintaining high-quality reconstructions.
Researchers developed a convolutional neural network to identify structures in cryo-X-Ray-microscopy data, achieving high accuracy within minutes. The AI-based analysis method enables faster evaluation of 3D X-ray data sets and has potential applications in studying cell responses to environmental influences.
EPFL researchers have created a novel biosensor, ImmunoSEIRA, to detect misfolded protein biomarkers linked to Parkinson's and Alzheimer's diseases. The sensor employs AI-powered neural networks for disease stage quantification and features gold nanorod arrays with antibodies for specific protein detection.
Developing a technique to create conductive polymer wire connections between electrodes enables artificial neural networks that overcome the limits of traditional computer hardware. The approach allows researchers to control and train the network using small voltage pulses.
Researchers at EPFL have found a way to teach quantum computers to learn and process information using principles inspired by quantum mechanics. By training quantum neural networks (QNNs) on a few simple examples called 'product states', the computer can effectively grasp complex dynamics of entangled quantum systems.
A team of researchers from Kyoto University and international institutions has developed a mathematical solution to the temporal asymmetry of nonequilibrium disordered Ising networks. This breakthrough offers insights into the behavior of biological systems, machine learning, and AI tools.
Researchers summarize existing compiler technologies in deep learning co-design and propose a new framework, the Buddy Compiler, to address performance bottlenecks in current AI applications. The study highlights the importance of hardware-software co-design in achieving optimal efficiency and effectiveness in deep learning systems.
A new AI-based technique measures brain fluid flow with unprecedented accuracy, revealing pressures and three-dimensional flow rates. This breakthrough could lead to the development of new treatments for Alzheimer's, small vessel disease, strokes, and traumatic brain injuries.
A new approach to enhance artificial intelligence-powered computer vision technologies has been developed by UCLA researchers, adding physics-based awareness to data-driven techniques. This hybrid methodology aims to improve how AI-based machinery sense, interact, and respond to their environment in real time.
The NEHO project aims to create ultrafast and energy-efficient information processing systems using photonics and semiconductor technology. By leveraging nonlinear photon-plasmon interactions, researchers hope to revolutionize information processing with faster, more efficient, and flexible technologies.
Researchers at NYU Grossman School of Medicine have developed an AI tool called NYUTron that can accurately estimate patients' risk of death, length of hospital stay, and other factors important to care. The tool achieved impressive results in predicting readmission rates, improving upon standard methods by up to 7%.
EPFL researchers used Chat-GPT to design a working robotic tomato harvester, showcasing the AI tool's potential for collaborating with humans in robotic design. The study highlights opportunities and risks of applying artificial intelligence to robotics, emphasizing the need for careful evaluation of LLMs' role in design.
PolyU researchers have developed optoelectronic graded neurons that can perceive dynamic motion, achieving an information transmission rate of over 1000 bit/s. This breakthrough enables highly accurate motion recognition, surpassing conventional image sensors by up to 99.2% accuracy.
Researchers developed an AI algorithm called CathEF to estimate left ventricular ejection fraction (LVEF) from standard angiogram videos, providing real-time information for clinical decision-making. The tool was trained on a large dataset and demonstrated strong correlations with echocardiographic LVEF measurements.
A team of researchers developed an unsupervised entity alignment framework to improve knowledge graph search, avoiding human labor. The framework outperformed most competitors on precision and recall, scoring higher overall across multiple datasets.
Researchers develop a new online learning algorithm that enables the training of larger spiking neural networks with six million neurons. This allows for faster and more efficient processing of tasks such as speech recognition and object detection.
Researchers have developed a new method called EvoAug that uses artificial DNA sequences inspired by evolution to train deep neural networks for genome analysis. This approach enables the model to recognize regulatory motifs more accurately, leading to better performance and potential breakthroughs in understanding human health.
The UMD-led TRAILS institute will develop AI technologies that promote trust and mitigate risks through broader participation, new technology development, and informed governance. The institute aims to create AI systems that align with values and interests of diverse groups, leading to increased transparency, reliability, and accountab...
A deep neural network developed by researchers at the University of California - Santa Cruz has been shown to accurately classify particle signals with 99.8% accuracy in real-time. The system can identify weak or noisy signals and pinpoint their source, making it suitable for point-of-care applications.
Researchers at the University of Pennsylvania School of Engineering and Applied Science have created a photonic device that provides programmable on-chip information processing without lithography. This breakthrough enables superior accuracy and flexibility for AI applications, overcoming limitations of traditional electronic systems.
Researchers at the University of Texas at Austin developed a semantic decoder that translates brain activity into text, allowing individuals with speech disabilities to communicate. The system has been trained on extensive hours of podcasts and can decode continuous language, capturing the gist of what is being said or thought.
Researchers at Sainsbury Wellcome Centre found that instinctual exploratory runs enable mice to learn a map of the world efficiently. The study demonstrates how biological brains can learn faster and more efficiently than AI agents by focusing on salient objects.
Researchers from Osaka University developed an AI algorithm called FINDE that discovers and preserves the underlying conservation laws of real-world dynamical systems, not just superficial dynamics. FINDE allows for more accurate computer simulations and can reveal additional information about a system's structure.
A new neural network, CD-GAN, uses common sense knowledge to enhance text descriptions and generate images of birds at three resolution levels. The system achieved competitive scores against other image generation methods, producing vivid and natural-looking images.
Researchers found that simultaneous learning of two tasks enhances a deep-learning model's performance on retrieving precipitation information from satellite data. The new framework uses multi-task learning to improve current estimates of precipitation, outperforming existing approaches.
Researchers developed a high-speed prediction model combining physical simulations and machine learning, achieving high accuracy without compromising computation time. The technology uses correspondence between input physical conditions and abstract data space handled by machine learning algorithms.
Researchers at King Abdullah University of Science & Technology (KAUST) successfully integrated two-dimensional materials on silicon microchips, achieving high integration density, electronic performance, and yield. The resulting hybrid devices exhibit special electronic properties that enable low-power consumption artificial neural ne...
A University of Illinois project uses AI-powered object recognition to quantify kernel damage in wheat, enabling faster disease analysis and improved resistance. The technology has shown promising results, with potential for an online portal to automate scoring and support breeders in their efforts to eliminate fusarium head blight.
Scripps Oceanography researchers developed a machine learning method to separate fish chorusing sounds from the overall ocean noise, enabling faster analysis and identification. The 'SoundScape Learning' technique can be applied to other soundscapes to learn more about animals like frogs, birds, and bats.
Researchers developed a deep learning-based AI model to automate cirrhosis identification using large amounts of data from EHRs. The model successfully identified patients with cirrhosis with a precision of 97%, offering potential for early diagnosis and improved management of the disease.
Researchers developed AI models based on UNet and MobileNet architectures to analyze standardized abnormalities in CT images, accurately identifying object presence and confidence. These models achieved an absolute percentage error of less than 5 percent, comparable to human professionals.
Researchers at Stanford University have developed a novel AI-powered approach to analyzing traumatic brain injury, using artificial intelligence to identify the most accurate model of mechanical stress on the brain. This breakthrough could lead to better understanding of when concussions lead to lasting brain damage and inspire new pro...
Researchers seek to develop algorithms providing meaningful explanations for AI decision-making, enabling higher human trust and adoption in fields like science. The project focuses on symbolic reasoning and estimating explanation accuracy, addressing the need for transparent AI systems.
GPT-3 performs nearly on par with humans in decision-making but struggles with causal reasoning and information search. The language model's limitations may be due to its passive information-gathering approach, highlighting the need for active interaction with the world to achieve human-like intelligence.
A POSTECH research team has developed a deep-learning approach to enhance resolution and speed in photoacoustic computed tomography (PACT) imaging. The technique enables high-resolution, real-time whole-body imaging of animals and monitors tissue movement in the heart, kidney, and brain.
Researchers at Ruhr University Bochum developed a new digital imaging method using artificial intelligence and infrared imaging to determine microsatellite status in colon cancer. This approach enables fast, label-free, and automated detection of the biomarker, which is crucial for personalized medicine.
A new study uses Fourier analysis to understand how deep neural networks learn complex physics. By analyzing the equation of a fully trained model, researchers were able to identify crucial information about how the network learns and generalizes. This breakthrough could accelerate the use of scientific deep learning in climate science.
Researchers developed a neural network model that uses terahertz time-domain spectroscopy data to predict burn healing outcomes with high accuracy. The new approach improves upon existing methods by reducing training data requirements, making it more practical for processing large clinical trials.
Researchers at KAIST developed a quadrupedal robot control technology that enables robots to walk robustly on deformable terrain like sandy beaches. The technology uses artificial neural networks to simulate ground characteristics and adapt to changing environments, allowing the robot to maintain balance and perform high-speed walking.
A new system called EasySort AUTO uses artificial intelligence image recognition to sort single bacterial cells, increasing efficiency by over 93% and preserving cell vitality. The technology has been successfully tested on yeast and E. coli bacteria, with over 80% of sorted cells able to be cultured.
Researchers from Brown and MIT developed a new framework that uses machine learning and sequential sampling to predict rare disasters like earthquakes and pandemics with less data. The framework, called DeepOnet, has been shown to outperform traditional modeling efforts in predicting scenarios, probabilities and timelines of rare events.
Researchers developed an AI-based neural network to detect early knee osteoarthritis from x-ray images, matching doctors' diagnoses in 87% of cases. This method could help reduce unnecessary examinations, treatments, and even knee joint replacement surgery.
A team of researchers from the University of Pennsylvania has developed a new algorithm, metadynamics, that can navigate high-dimensional energy landscapes to find low-energy configurations. This breakthrough has the potential to revolutionize fields such as protein folding and machine learning.
A neural network trained using a diverse dataset outperforms conventionally trained algorithms by reducing bias in artificial intelligence. The use of images from low-resource populations boosts the object recognition performance of machine learning systems.
Engineers at Tokyo Tech demonstrated a simple approach to improve AI classifier training using limited sensor data, increasing quality without extra cost. The proposed method promises to address the challenge of classification accuracy in real-world applications, where reliable answers are crucial.
Researchers from the University of Johannesburg deployed Few Shot Learning (FSL) for NIALM, a non-intrusive appliance load monitoring system. FSL requires only 7 test images to recognize appliances with 97.83% accuracy, making it faster and more cost-effective than traditional Machine Learning.
Researchers found that artificial neural networks learn better with occasional periods of inactivity, similar to the human brain's benefits of sleep. This approach mitigates catastrophic forgetting and allows for continuous learning, like humans or animals.
Researchers created synthetic knee x-ray images to complement real images in osteoarthritis classification. Medical experts were unable to distinguish between authentic and synthetic images, highlighting the potential of synthetic data for collaboration and testing.
Researchers induced brain-like sleep in an artificial spiking neural network, enabling it to retain previously learned information. This breakthrough helps the model avoid catastrophic forgetting and improve its overall performance over time.
The AlphaFold2 AI model has contributed 25% more high-quality protein structures to existing species, aiding in understanding protein function and designing targeted drugs for cancer. Despite limitations, its impact will transform life sciences with new computational tools.
Researchers trained a mouse brain model to solve visual tasks, achieving superior robustness and performance compared to traditional neural networks. The model's unique coding properties enable it to cope with errors and unexpected input, making it a promising tool for advances in neuromorphic computing.
Deep learning models can become less accurate in recognizing specific categories of images, sounds, or text after network pruning. Researchers demonstrate a technique to address this challenge, improving the fairness of deep learning models.